"Beyond GridPP2 and e-Infrastructure" ===================================== Panel Members: Pete Clarke, Dave Britton, Tony Doyle, Neil Geddes, John Gordon, Neasan O'Neill, Joanna Schmidt, John Walsh, Pete Watkins. Introductions ============= John Gordon: Most important is the integration with NGS . We have a window of opportunity as we are similar at the moment. NGS will transfer to GT4, multiple stacks e.g. glite. Need to get VO's accepted with NGS. NGS has concept of VO's but not implementation. Then we will have a UK grid - in a good position for a euro grid. Joanna Schmidt: from White Rose Grid. Support JG that interoperability is most important. Need and integrated approach within UK and then Europe. John Walsh: from Grid Ireland, will give Irish perspective. NG? discussion paper - a problem is that we have short term projects. Next stage of evolution needs longer term support EDG and data-grid focused on the physics domain. EGEE and NA4 changed a lot - more pervasive - NGS does this in the UK. NRENS - National Resource centres model through europe - works very well, point of prescence in every university. Same with grid middleware. Needs to be easy to put in place. Pete Watkins: from Birmingham. This is a tricky time at the moment. We have lots of Gridpp problems to solve. However, will have no time once LHC starts up to do peripheral stuff. e-science and SRIF3 were big investments; we need to make more easy for the typical university researcher to submit a job. Big challenge to integrate into JISC long term. Support every effort before LHC switch on to bring the grid together. Neil Geddes: Emerging ways of standardising e-infrastructure. Not just about submitting jobs. There is lots of industrial hype. Standard interaces (Google?) based on http, soap, web services. Do we do enough as a community to influence these standards or do we do our own thing? This is the risk to mitigate against. Depends on likelyhood of Gridpp3 being funded. Pete Clarke: What is e-science - no idea! The future revolves around being good citizens nationally. We don't integrate enough. Need to use JISC more, could do more. New PPARC needs to be lobbied to be more integrated. Gridpp/NGS very important politically. Even if painful it is worth the pain to integrate. Need to make GridPP part of NGS - big benefit in the long run. Need tentacles in universities also. Running computers is now visible because of costs of electricity. Historically wedded to CPU and disks. Need to lose this and need to focus on services. Tony Doyle: National e-infrastructure. 10 year Tony Hay, OMII software engineering, UK repository - hasn't really happened. Pressures will force this to converge. Digital curation will be important. Data from OPAL only a terabyte, LHC will be a petabyte and will be analysed for a long time in future. On multi-dsciplinary working: need to attend all hands meeting and GGF - we did more in past, but now focus on our own problems. Dave Britton : How do we see the future without GridPP? Once we know this then we will know how to get there. NGS, but what about Tier-2's without GridPP what should we do now for this? Open discussion =============== James Werner: BNP Paribas are using grid techniques. This is a good opportunity for jobs - maylose people at LHC start up. Valuable people are ones who can get things to work. John Gordon: GridPP developed these people. Need focus to keep them involved for the next project; whether a community led or external project. Dave Britton: The people interested in development will move on - they want new challenges. Training is our role also - part of the game. John Walsh: Some governments would like to see this (movement of people). Tony Doyle: People will come to LCG once LHC starts up because it will become more interesting. Pete Gronbach: I agree, people work in GridPP because they are interested in it. John Gordon: People might be let down by industry's version of the grid - basically it's CPU scavenging in a single domain. Jeremy Coles: I don't see an individual whose responsibility it is to bring things together. Tony Doyle: Tony Hay will be on the new large facilities council - he is a key person. Pete Watkins: I agree - someone in the project needs to have that as their responsibility. Pete Clarke : RAL lead on this already. discussion on large university computing facilities.... Robin Middleton: Can I come back to the Tier-2 model. What is the model for the future? Pete Clarke: Future computing service - take a lead for PPARC experiments - like Tier-1. As for Tier-2 too much interest in CPUs and terabytes. Tier-2s should focus on services. A university needs to decide if it needs software services for the grid, when ratio of hardware to staff becomes so then it will do it. Andrew McNab: Money is tied to projects. Neil Geddes: Business will buy services if cheaper than starting up themselves. For particle physics it is cheaper not to outsource because we have very tight specifications. Tony Doyle: Scientific Linux is now becoming a standard. Paul Miller: Move towards virtualisation? Making it simple is the way forward. One way is to a run a virtual environment. Mark Nelson: I agree. Large data production. If put it across BT/commercial networks will cost a lot. John Gordon: No don't agree. Pete Clarke: Question is what direction will country go in. JISC might outsource computing services. One last point: telco in UK is outdated. Problem with CPU cost is it is aimed at banks - high costs. Paul Miller: What about amazon S3? John Walsh: Can buy compute services and storage but how does collaboration work? Pete Clarke : Perhaps we should try it. John Walsh: EGEE see scientific linux as a problem; non-portability is a problem - nightly build results are poor. Neil Geddes: Biomed community like Microsoft Windows. There is a risk in our divergence. Pete Clarke: Correction: Tony Hay is on PPARC council, not yet on large facilities council. Robin Middleton: What about the european dimension? Is it a no-brainer that we remain aligned? Neil Geddes: EGEE will fund 14 people at CERN - they are aligned. Robin Middleton: As EGEE moves out from particle physics other VO's will become major players. Neil Geddes: We are ahead of other european countries. Other VO's not so interested in high throughput. Need to stay engaged. Pete Gronbech: Broadening out has given us valuable publicity in terms of getting funding from government. Pete Watkins: Not many VO's are of UK origin (except pheno). Robin Middleton: What can we do about this? Neil Geddes: Most other countries get money from Europe via matched funds. UK is relatively rich. Need to be more proactive in the UK. John Gordon: There are other functioning grids about the UK e.g. e-minerals but they have a single focus. Robin Middleton: Are there any other opportunities? Tony Doyle: There is a big event in May in Manchester. It is a joint meeting between EGEE and OGF. Large international body coming to UK. Easy to attend and be influenced by discussions. Robin Middleton: It is a golden opportunity for GridPP. Pete Clarke: It will be a self-selecting audience. No overlap with other UK researchers. There are call outs from JISC. How can we move our research such that it can be used by others? Joanna Schmidt: Pleased to see JISC suporting e-science. Jens Jensen: We can engage with other countries. The german grid is the driving power for their e-science program. They will have similar problems. Neil Geddes: Coming back to the May event, we really need non particle physicists to go to the EGEE user forum. We are unlikely to have someone put in charge. Pete Clarke: There is a call for virtual research environments - this is reinventing what we have done for a long time. They are not calling on particle physicists to contribute to this. John Gordon: This could be a possible source of funding for the Imperial portal. Robin Middleton: What about web services, where should we be going with these? John Gordon: We are using web services: FTS, LFC, SRM are all web services. Paul Miller: 'Restful interface' is an alternative approach. We don't want to be dictating a particular technology. e.g. google provide a number of api's, calendering is now rest based. Andrew McNab: These are still web services. Jens Jensen: We must use standards e.g. SRM is using plain old web services. Paul Miller: Coming back to cost of out sourcing with amazon S3 it is £10k per year to store 10TB. Pete Clarke: That is good value. Dave Britton: For Gridpp3 it is £128 per TB-year for hardware. Must add on extras. John Walsh: For the users forum: is GridPP best organisation to encourage other communities? Wouldn't NeSC be better placed? John Gordon: We should all be doing it. Pete Clarke: NeSC is doing this. Neil Geddes: Collaboration is the key factor.